Bloated Disclosures: Can ChatGPT Help Investors Process Financial Information?

Alex Kim, Accounting PhD student
Maximilian Muhn, Professor of Accounting
Valeri Nikolaev, Professor of Accounting and FMC Faculty Scholar

Generative AI tools such as ChatGPT could fundamentally alter the way financial markets process information. We plan to probe the value of generative AI in extracting useful textual information from a large set of unstructured corporate disclosures. Specifically, we plan to use the GPT model to summarize the information content in MD&A and earnings call transcripts, to create a novel measure for the degree of redundant information and to explore standardized theme-specific summaries. Using the GPT-3.5 model on a small 20% sample, we found promising early results. The results suggest that unconstrained summaries are dramatically shorter, whereas their information content is amplified. For example, summarized sentiment is considerably more effective at explaining stock market reactions to the disclosed information than the original. Motivated by these findings, we propose a novel measure of the degree of redundant disclosure (“Bloat”). Our early findings suggest that bloated disclosures are associated with adverse capital market consequences (e.g., higher information asymmetry).